PLCγ1 promotes phase separation of the T cell signaling clusters

The T cell receptor (TCR) pathway receives, processes, and amplifies the signal from pathogenic antigens to the activation of T cells. Although major components in this pathway have been identified, the knowledge on how individual components cooperate to effectively transduce signals remains limited. Phase separation emerges as a biophysical principle in organizing signaling molecules into liquid-like condensates. Here we report that phospholipase PLCγ1 promotes phase separation of LAT, a key adaptor protein in the TCR pathway. PLCγ1 directly crosslinks LAT through its two SH2 domains. PLCγ1 also protects LAT from dephosphorylation by the phosphatase CD45 and promotes LAT-dependent ERK and SLP76 activation. Intriguingly, a non-monotonic effect of PLCγ1 on LAT clustering was discovered. Computer simulations, based on patchy particles, revealed how the cluster size is regulated by protein compositions. Together, these results define a critical function of PLCγ1 in promoting phase separation of the LAT complex and TCR signal transduction.


Introduction
Mesoscale signaling clusters have been frequently observed in a variety of immune receptor pathways, including the T cell receptor (TCR) (Campi et al., 2005), B cell receptor (BCR) (Wang et al., 2017), Fc gamma receptor , engulfment receptor (Draper) (Williamson and Vale, 2018), T cell co-receptor CD28 (Yokosuka et al., 2008), and PD1 (Hui et al., 2017). These signaling clusters, ranging from several hundred nanometers to microns in size, share common features of complex composition, heterogeneous size, and dynamic assembly (Dustin and Groves, 2012). Because of these features, it remains a challenge to understand the mechanism and functional consequences of those clusters in transducing immune signaling.
The T cell microcluster represents a good example of this scenario. Following TCR activation, downstream signaling molecules self-organize into micron or submicron-sized clusters. These clusters are enriched of TCR, adaptor proteins LAT, Grb2, Gads, SLP76, Nck, and effectors ZAP70, Sos1, PLC1, CBL, and WASP (Balagopalan et al., 2015;Bunnell, 2010;Bunnell et al., 2002). Because the majority of early TCR signaling components reside in these microclusters, they are considered as a hub for transducing TCR signals (Balagopalan et al., 2015;Choudhuri and Dustin, 2010). Previous works showed that LAT, a transmembrane protein essential for TCR signal transduction (Zhang et al., 1998), serves as a scaffold to form a macromolecular signaling complex (Houtman et al., 2006;Zhang et al., 2000). Mathematical modeling suggested that multivalent protein-protein interactions play a critical role in forming the LAT complex (Nag et al., 2012;Nag et al., 2009). Using a supported lipid bilayer-based system, our previous work showed that LAT forms near micron-sized, membrane-embedded clusters through a mechanism of liquid-liquid phase separation (Su et al., 2016). The liquid-like LAT microclusters enrich kinase ZAP70 but exclude phosphatase CD45, thus promoting tyrosine phosphorylation. LAT microclusters also increase the downstream Ras activation , actin polymerization, and ERK activation (Su et al., 2016). These works revealed a critical role of LAT microclusters in promoting TCR signaling. However, the regulatory mechanism of LAT microclusters was not fully understood.
PLC1 is a multi-domain lipase. Besides its catalytic core, PLC1 also contains two SH2 domains and one SH3 domain that serve structural or regulatory roles Manna et al., 2018). Its N-terminal SH2 domain (nSH2) binds specifically to phosphor-tyrosine (Y132) on LAT  whereas its C-terminal SH2 domain (cSH2) is involved in releasing autoinhibition of PLC1 (Gresset et al., 2010;Hajicek et al., 2019). The SH3 domain of PLC1 directly interacts with the proline-rich motifs on Sos1 (Kim et al., 2000), a RasGEF which is also enriched in the LAT microclusters. The multiple binary interactions between PLC1 and other components in the LAT complex mediate a synergistic assembly of the LAT complex Hartgroves et al., 2003;Manna et al., 2018). However, because of the complex protein-protein interactions involved, the exact mechanism of how PLC1 regulates LAT microcluster formation remains elusive. This is mainly because traditional biochemical assays were performed in solution, which did not recapitulate two important features of cellular LAT microclusters: membrane association and giant size (up to micron).
We have recently developed a supported lipid bilayer-based system to reconstitute nearmicron-sized LAT microclusters on synthetic membranes . In the following study, using our biochemical reconstitution approach together with live cell microscopy and computer modeling, we delineated the role of PLC1 in regulating LAT microcluster formation. We found that the SH2 and SH3 domain of PLC1 can directly bridge LAT to Sos1 to form microclusters. PLC1 also protects LAT from CD45-dependent dephosphorylation. Therefore, PLC1 stabilizes LAT microclusters by both physical crosslinking and chemical protection. Moreover, we found that the PLC1 concentration influences the sizes of LAT microclusters in a non-monotonic way, pointing to a novel mechanism of the size control of liquid condensates. Together, these results expand the traditional view that PLC1 acts downstream of LAT; instead, PLC1 plays an active role in regulating the stability of LAT microclusters.

PLC1 promotes LAT cluster formation in vitro
To determine how PLC1 regulates LAT microcluster formation, we implemented a supported lipid bilayer-based reconstitution assay that allows quantitative monitoring of microcluster assembly . The cytoplasmic domain of LAT was purified, phosphorylated, and labeled with maleimide-Alexa488 on a C-terminal cysteine residue.
It was then attached to the Ni2 + -NTA functionalized supported lipid bilayer via a polyhistidine tag on the N-terminus. Total internal reflection fluorescence (TIRF) microscopy was used to visualize the formation of LAT clusters. As reported before (Su et al., 2016), LAT formed microclusters when Grb2 and Sos1 were added to the reaction mixture ( Figure 1A and 1B). Intriguingly, when Grb2 was replaced with a fragment of PLC1 that contains the SH2 and SH3 domains, LAT still formed clusters, though with a higher number but smaller sizes, as compared to the Grb2-mediated cluster formation ( Figure 1B). Furthermore, fluorescence recovery after photobleaching (FRAP) analysis revealed that PLC1-induced LAT clusters are less dynamic than Grb2-induced LAT clusters, as indicated by the lower recovery frequency and longer half recover time ( Figure 1C). We also tested the full-length PLC1 and found that it induced LAT microcluster formation in a dose-dependent manner, which is similar to the fragment of PLC1 ( Figure S1A and S1B). To understand if the ability to drive LAT cluster formation is a general feature of proteins containing the SH2 and SH3 domains, we replaced PLC1 with other LAT-associated proteins that play a role in TCR signal transduction. These include 1) Gads, an adaptor protein that binds LAT on overlapping sites with Grb2 (Zhang et al., 2000), 2) Vav1, a RhoGEF that closely associates with LAT (Sherman et al., 2016), and 3) Nck1, an adaptor protein that promotes actin polymerization (Wunderlich et al., 1999). We found that Gads, as reported before (Su et al., 2016), drives LAT microcluster formation whereas Vav1 or Nck1 did not ( Figure S1C and S1D). Together, those data showed that PLC1, together with Sos1, can specifically induce LAT microcluster formation.

PLC1 crosslinks LAT through two SH2 domains
Next, we determined the mechanism by which PLC1 drives LAT microcluster formation. PLC1 contains an N-terminal SH2 domain (nSH2), a C-terminal SH2 domain (cSH2), and an SH3 domain. We produced PLC1 truncation mutants that lack either the nSH2, cSH2, or SH3 domain ( Figure 2A). We found that mutants lacking either nSH2 or cSH2 lost the ability to drive LAT cluster formation whereas mutants lacking the SH3 domain still drove LAT cluster formation ( Figure 2B and 2C). Because the SH3 domain interacts with the proline-rich motif on Sos1, the SH3-independent clustering suggested that Sos1 might be dispensable for PLC1-driven LAT cluster formation. Indeed, the nSH2-cSH2 fragment of PLC1 drove LAT cluster formation in the absence of Sos1 ( Figure S2A and S2B). In this assay, PLC1 and mutants were used at 500 nM. We titrated PLC1 concentration and found that at a lower concentration of PLC1 (50 nM), Sos1 is still required for LAT microcluster formation ( Figure S2C and S2D).
Next, we determined the binding sites on LAT that interact with the SH2 domain of PLC1.
The four C-terminal tyrosines on LAT are necessary and sufficient to transduce the TCR signaling (Zhu et al., 2003). Synthesized peptides containing each one of these four phosphotyrosines of LAT were attached to the supported lipid bilayer. Recombinant nSH2 or cSH2 domain of PLC1 was purified, labeled with CoA-647 on an N-terminal ybbR tag, and incubated with individual phospho-peptides on the membrane ( Figure 2D). The binding of the SH2 domain to phosphopeptide on the membrane was revealed by TIRF microscopy. We found that the nSH2 domain strongly interacted with LAT Y132 ( Figure   2E), which is consistent with previous reports (Zhang et al., 2000). The cSH2 domain robustly bound to LAT Y171, though the binding was slightly lower than PLC1 Y783 ( Figure 2F), the previously reported site that interacts with the cSH2 domain in cis (Hajicek et al., 2013;Poulin et al., 2005). Together, those data suggested a model in which PLC1 crosslinks LAT through two interaction pairs: nSH2 with Y132, and cSH2 with Y171.

PLC1 cooperates with Grb2 to regulate LAT clustering
Next, we investigated how PLC1 cooperates with Grb2 and Sos1 to induce LAT microcluster formation. Because cluster formation depends on the concentration of individual components, we adopted concentrations of proteins that were measured in T cells (Nag et al., 2009;Voisinne et al., 2019). We incubated LAT at a density of 300 molecules / m 2 with 3000 nM Grb2 and 300 nM Sos1. PLC1 was additionally included in the clustering assay. Surprisingly, we found that PLC1, at 50 nM (concentration in T cells), could significantly increase LAT cluster formation and recruitment of Sos1 to the membrane. This clustering-promoting effect is robust even at a concentration as low as 5 nM PLC1 ( Figure 3A and 3B). Notably, this concentration (5 nM) is orders of magnitude lower than Grb2 (3000 nM), the other SH2-SH3 adaptor in the system. Furthermore, FRAP analysis revealed that PLC1 significantly reduces the exchange of LAT molecules between inside and outside of the clusters ( Figure S3), supporting the idea that PLC1 serves as an orthogonal crosslinker to Grb2, to stabilize LAT clusters. To understand how PLC1 affects the kinetics of LAT cluster formation, we performed time lapse imaging.
LAT was attached to the membrane, Grb2/Sos1 was added into the system at time 0. We observed cluster formation as usual ( Figure 3C and Movie S1). Intriguingly, when PLC1 was additionally included in the system, clustering was significantly enhanced from the very beginning ( Figure 3C and Movie S2). Together, these data suggest that, PLC1, by serving as a crosslinker, could dramatically increase LAT microcluster formation.

Computer Simulations of PLC1-mediated LAT clustering
Intriguingly, a non-monotonic effect of PLC1 on LAT clustering was revealed. PLC1, at low concentrations, promotes LAT clustering and increases cluster sizes but this effect is diminished at high concentrations ( Figure 3A and 3B). This points to the fact that PLC1 not only promotes LAT clustering, but also regulates the cluster size. To understand the physical mechanism underlying this size regulation, we developed a minimal coarsegrained computer model in which the proteins are described as spherical particles decorated with patches that represent binding domains and simulated via molecular dynamics. The model, described quantitatively in Methods and in Supplemental Information (SI), consists of 4 types of two-dimensional particles, representing LAT, PLC1, Sos1 and Grb2; these can bind mutually, respecting biochemical valence and bond specificity ( Figure 4A). In simulations, the pool of such two-dimensional particles readily aggregates into clusters that grow either via addition of individual proteins or via merging with other clusters until they reach steady-state sizes. The average cluster size in simulations exhibits a non-monotonic dependence on the concentration of PLC1 ( Figure 4B, and Movie S3-S8), indeed recapitulating what was observed in experiments ( Figure 3B). The non-monotonic effect was robustly revealed in a wide range of LAT densities with cluster size measured either by the number of LAT or by the total number of four proteins ( Figure S4A). To understand the origin of this non-monotonic phenomenon, we computed, at a given time, the number of possible bonds between free binding sites that can make whichever two clusters merge into a bigger one; this parameter, that we named coalescence likelihood, is a measure of how easy it is for the clusters to merge at a certain PLC1 concentration. We found that the coalescence likelihood closely captures the non-monotonicity of the cluster size ( Figure 4B). A breakdown of the coalescence likelihood by bond type (Figure 4C), together with an analysis of the average coordination per molecule (Section SI 3), showed the following.
At low concentrations, PLC1 particles provide a binding site for the otherwise unbound pY132 in LAT particles and increase binding of the pY171 site and of all Sos1 sites, through an SH3-PRM (proline-rich motif) bond. The availability of these additional binding sites increases the coalescence likelihood (yellow and orange bars in Figure 4C); at the same time, it increases the average coordination (number of sites that interact with other molecules) of both LAT and Sos1 particles, making clusters more connected ( Figure 4E and S4B). At high concentrations, though, PLC1 ends up saturating PRMs on Sos1 and, more slowly, phosphotyrosines on LAT particles, with the help of Grb2 ( Figure S4B). As this occurs, free binding sites become rare and the coalescence likelihood decreases drastically. Clusters are still more connected and compact than without PLC1, but they are smaller: indeed, Sos1 and LAT are almost fully bound and it is unlikely for new bonds to form upon random collisions (pink and gray bands in Figure 4C restrict by more than half, Movie S8).
A further analysis on the compactness of clusters was performed by computing their inertia tensor, as well as by graph-theoretical means (see Methods and Figure S4B). The compactness index showed that clusters become monotonically more compact as PLC1 concentration increases ( Figure 4D). Nonetheless, at high concentrations, clusters exhibit a large amount of 'terminal nodes', i.e. particles bound to one other particle only. This is due to the overwhelming amount of PLC1 particles, which tend to cap any binding site available to them, competing with themselves and with Grb2 for LAT and Sos1 sites. At the same time, as expected, LAT and Sos1 molecules become more and more bound ( Figure 4E and S4B). This suggests that the stabilizing effect of PLC1 observed in FRAP experiments is due to increased compactness of clusters and increased complexity of the LAT network therein. In short, PLC1 concentration emerges as a possible regulator both of cluster size and of cluster stability.

PLC1 promotes LAT clustering in T cells
We then investigated how PLC1 regulates LAT cluster formation in T cells. A LAT-mCherry construct was introduced into the wild-type or PLC1 null Jurkat T cells by lentiviral transduction. Those T cells were activated by cover glass-coated TCR-activating antibody OKT3. The formation of LAT microclusters was monitored by TIRF microscopy.
We found that LAT clustering in PLC1 null cells was significantly reduced as compared to the wild-type cells ( Figure 5A). This suggested a positive role of PLC1 in promoting LAT clusters, which is consistent with forehead in vitro results. To understand how the SH2 and SH3 domain of PLC1 contribute to LAT clustering, we reconstituted PLC1 null cells with the full-length PLC1 or PLC1 lacking the nSH2, or SH3 domain. These cells also express LAT-mCherry, as an indicator for LAT clusters. Unfortunately, PLC1 null cells reconstituted with PLC1 lacking the cSH2 domain do not grow, potentially because of the hyperactivity and toxicity resulting from the deletion of the inhibitory cSH2 domain.
We found that LAT microcluster formation was significantly higher in the wild-type cells, as compared to the nSH2 or SH3 cells ( Figure 5B).
Because LAT clustering activates downstream signaling pathway including SLP76 and MAPK, we decided to determine if PLC1 affects these pathways. Jurkat T cells were activated by anti-CD3 and anti-CD28 antibodies before being harvested for immunoblot analysis. Indeed, as compared to cells expressing the wild-type PLC1, cells expressing PLC1nSH2 or PLC1SH3 displayed defects in SLP76 activation (as indicated by SLP76 pY145) and MAPK activation (as indicated by pERK1/2) ( Figure 5C and 5D). We will discuss the possible pathway leading to these defects in Discussion. Together, these data suggest that the structural domains of PLC1 promote LAT clustering and LAT downstream pathways in T cells.

PLC1 protects LAT from dephosphorylation
The formation of LAT microcluster is phosphorylation-dependent. Antigen engagement with TCR triggers the phosphorylation of LAT by the kinase ZAP70, which is antagonized by phosphatases (Su et al., 2016). LAT, once phosphorylated, can recruit SH2-containing proteins to form microclusters. Therefore, PLC1 has traditionally been viewed as a "passenger" that is passively recruited to phosphorylated LAT. Interestingly, we found that PLC1 also regulates the phosphorylation of LAT. As compared to the wild-type cells, PLC1 null cells had significantly lower phosphorylation on Y132, but not on Y191 ( Figure   6A), a binding site for Grb2 and Gads (Zhang et al., 2000). Deleting Grb2 or Gads, the other major SH2 domain-containing proteins in the LAT complex, did not alter the phosphorylation of Y132 ( Figure 6A). This is consistent with the fact that Y132 only binds PLC1 but not Grb2 or Gads. Intriguingly, overexpressing a fragment of PLC1 that contains the SH2 and SH3 domains significantly increased phosphorylation on Y132 ( Figure 6A), suggesting PLC1 as a two-way (both up and down) regulator of Y132 phosphorylation. Because PLC1 is the only identified binding partner of LAT Y132, we hypothesized that the SH2 domain of PLC1 binds to LAT Y132, protecting it from being dephosphorylated. Supporting that, when the cellular phosphatase activity was inhibited by vanadate, the difference in phosphorylation level of Y132 between the wild-type and PLC1 null cells was abolished ( Figure 6B). To directly test the hypothesis that PLC1 protects LAT Y132 from dephosphorylation, we set up an in vitro dephosphorylation assay.
LAT-Grb2-Sos1 microclusters were assembled in the presence or absence of PLC1.
Then CD45, the most abundant phosphatase on T cell membranes, was added to dephosphorylate LAT ( Figure 6C). Intriguingly, PLC1 significantly suppressed the dephosphorylation of LAT Y132 ( Figure 6D). Together, these data suggest that PLC1 specifically stabilizes the phosphorylation on LAT Y132 by protecting it from being dephosphorylated by CD45.

Discussion
PLC1 was traditionally considered as an enzyme that acts downstream of LAT. Following TCR activation, PLC1 is recruited to LAT microclusters and activated by Itk1-triggered phosphorylation. Activated PLC1 hydrolyzes PIP2 to generate IP3 and DAG, which triggers downstream calcium and PKC pathway, respectively. Here we reported that PLC1 can also promote LAT cluster formation. This is achieved by two ways: PLC1 crosslinks LAT directly or indirectly through Sos1; and PLC1 protects LAT from being dephosphorylated by CD45. Therefore, PLC1 and LAT reciprocally regulate each other in a positive manner.
The cSH2 domain of PLC1 is well-known for its role in regulating the enzymatic activity of PLC1 (Gresset et al., 2010;Hajicek et al., 2013). Combining our new results showing that cSH2 promotes LAT clustering, we propose the following refined model explaining the recruitment and activation of PLC1 during T cell activation ( Figure S6A): following TCR engagement with antigens, LAT is phosphorylated at Y171, Y132 and other sites.
These two sites interact with the cSH2 and nSH2 domain of PLC1, respectively, to recruit PLC1 from the cytosol to the membrane. Then together with Grb2, Sos1, and other LAT binding partners, PLC1 promotes LAT cluster formation. This will further recruit Itk1, which was shown to bind the LAT complex (Bunnell et al., 2000;Ching et al., 2000), to the LAT microclusters. Itk1 phosphorylates PLC1 on Y783 (Perez-Villar and Kanner, 1999). Phosphorylated Y783 interacts with the cSH2 domain intramolecularly to release the autoinhibition of PLC1 and activate the lipase activity. The cSH2 domain switches from LAT-binding to PLC1-binding, which reduces the avidity of PLC1 to the LAT complex. Consequently, PLC1 disassociates from the LAT clusters, either translocates to the TCR clusters (Cruz-Orcutt et al., 2014) or moves back to the cytosol.
Our domain truncation analysis of PLC1 in T cells revealed that the SH2 and SH3 domain could contribute to SLP76 and ERK activation. This could be explained by two nonexclusive mechanisms: 1) the SH3 domain of PLC1 directly binds Sos1 (Kim et al., 2000) and recruits Sos1 to the membrane, which facilitates Ras and ERK activation; similarly, the SH3 domain directly binds and recruits SLP76 to the membrane (Yablonski et al., 2001) in preparation for being phosphorylated by ZAP70; 2) the SH2 and SH3 domains promote LAT cluster formation, which further enhances the recruitment of Grb2 and Gads to LAT. Grb2 and Gads are constitutive binding partners for Sos1 and SLP76, respectively. Therefore, the SH2 and SH3 domain of PLC1 can contribute to downstream signaling both directly or indirectly through LAT.
Through both biochemical reconstitution and computational approaches, we revealed a non-monotonic mechanism of regulating LAT clustering by PLC1. Compositional control emerges as an important mechanism for regulating the physical feature and chemical activities of liquid-like condensates (Banani et al., 2016;Ditlev et al., 2019;Riback et al., 2020). The concentration of PLC1 may have been tuned to control cluster size and stability, with the purpose of transducing signaling to the physiological needs. Interestingly, PLC1 is upregulated in acute myeloid leukemia (Mahmud et al., 2017), colorectal carcinoma (Noh et al., 1994), and squamous cell carcinoma (Xie et al., 2010). It remains as an interesting question whether PLC1 mis-regulates signaling cluster formation in these pathological conditions. PLC1 is involved in a variety of membrane receptor signaling pathways. The natural killer cell and mast cell receptor (FcR1) pathway share almost the same machinery of LAT clustering with the T cell. We reasoned that PLC1 could play a similar role in promoting receptor signaling by enhancing LAT clusters in natural killer and mast cells. In the pathways outside immune responses, such as EGFR, FGFR or HER2, LAT is absent.
However, similar scaffold proteins are present that contain multivalent interaction sites (e.g. EGFR and Shc) that can interact with PLC1 and Grb2-Sos1. The cooperation between PLC1 and Grb2 in promoting receptor clustering, as revealed in this work, could serve as a general mechanism for regulating membrane receptor signaling.  (B) TOP: TIRF microscopy revealed that both Grb2 and PLC1 promote LAT microcluster formation. Alexa-488 labeled LAT at 300 molecules / μm 2 was incubated with 125 nM Sos1 and 250 nM Grb2 or PLC1 for 0.5 hr before imaging. Scale bar: 5 μm. BOTTOM: Quantification of Grb2 or PLC1 driven-LAT microclusters. LAT clustering was quantified as normalized variance (Su et al., 2016). Shown are mean ± SD. N= 3 independent experiments. Unpaired two-tail t-test was used. ***: p<0.001; ns: not significant.
(C) FRAP analysis revealed that PLC1 driven microclusters are less dynamic than Grb2driven LAT microclusters. Shown are mean ± SD. N= 10 clusters. (B) TIRF microscopy revealed that both nSH2 and cSH2 domains are required for PLC1 driven-LAT microcluster formation. SH3 domain promotes cluster formation. Alexa-488 labeled LAT at 300 molecules / m 2 was incubated with 300 nM Sos1 and 50 nM PLC1 for 0.5 hr before imaging. Scale bar: 5 m.
(C) Quantification of PLC1-driven LAT microclusters. Shown are mean ± SD. N= 3 independent experiments. Unpaired two-tail t-test was used. *: p<0.05 ; (D) Schematics of the assay of testing SH2 domain binding sites.
(E) PLC1 nSH2 binds LAT Y132. Phospho-peptides were synthesized, biotinylated at the N-terminus, and attached to the biotin-functionized supported lipid bilayers by streptavidin. The SH2 domain were labeled with fluorescent dye (Maleimide Ax647), and incubated with the individual phosphor-peptides. The membrane-associated SH2 domain was measured by TIRF microscopy. Scale bar: 5 m.

Figure 3 PLC1 cooperates with Grb2 to regulate LAT microcluster formation
(A) TIRF microscopy revealed that PLC1 regulates LAT microcluster formation in a nonmonotonic manner. Physiologically relevant concentrations of proteins were used in the assay: LAT at 300 molecules / m 2 , Grb2 at 3 M, Sos1 at 0.3 M, and PLC1 at 50 nM. LAT was labeled with Alexa-488, PLC1 was labeled with DY547, and Sos1 was labeled with Alexa-647. Scale bar: 5 m.
(C) PLC1 accelerates LAT cluster formation. TIRF microscopy revealed the time course of LAT microcluster formation in the presence or absence of PLC1. LAT-Alexa488, at 1000 molecules / m 2 was incubated with 1000 nM Grb2 and 500 nM Sos1, and/or  (C) Immunoblot analysis of LAT-null Jurkat T cells reconstituted with the wild-type, nSH2, or SH3 PLC1. Cells were stimulated with 2 μg/mL anti-CD3 and anti-CD28 antibodies for 2 min, lysed, and applied for western blot analysis.
(D) Quantification of the level of indicated proteins, after normalized to the expression level of GAPDH. Shown are mean ± SD. N= 3 independent experiments. Unpaired two-tail t-test was used. *p＜0.05, **p＜0.01.

Figure 6 PLC1 protects LAT from dephosphorylation by CD45
(A) Reduced phosphorylation at LAT Y132 in PLC1 null cells. Cells as indicated were stimulated with 2 μg/mL anti-CD3 and anti-CD28 antibodies for 2 min, lysed, and applied for western blot analysis. The level of indicated proteins, after normalized to the level of GAPDH, was quantified. Shown are mean ± SD. N= 3 independent experiments. Unpaired two-tail t-test was used. **p＜0.01.
(B) PLC1 prevents LAT Y132 from being dephosphorylated. Cells as indicated were pretreated with 0.1 mM vanadate (pan phosphatase inhibitor) before being stimulated with 2 μg/mL anti-CD3 and anti-CD28 antibodies for 2 min, lysed, and applied for western blot analysis. The level of indicated proteins, after normalized to the level of GAPDH, was quantified. Shown are mean ± SD. N= 3 independent experiments.
(C) Schematics of the in intro dephosphorylation assay.
The reaction was terminated by adding SDS-PAGE loading buffer with 2 mM vanadate.
The level of phosphorylated LAT, after normalized to total LAT, was quantified. Shown are mean ± SD. N= 3 independent experiments.

Recombinant Proteins
A panel of recombinant proteins used in this study was shown in Figure S6B.
The full-length PLC1 was purified from insect cells by a baculovirus expression system. PLC1 fragments containing individual or combined SH2 and SH3 domains were expressed and purified from bacteria. Peptides including each of the four LAT C-terminal phosphotyrosine residues (pY132, pY171, pY191, pY226) and PLC1 pY783 were synthesized by Peptide 2.0 Inc., Chantilly, VA. The peptides are modified with biotin at the N-terminus. The exact protein sequences of PLC1 and truncations, and synthesized peptides are listed in Table S1. Gads, Nck1, Grb2, Sos1, CD45, and LAT were purified as previously described (Su et al., 2016).

Protein purification
Full length PLC1 The full-length bovine PLC1 with an N-terminal His10 tag and a C-terminal SNAP tag was expressed in SF9 cells using the Bac-to-Bac baculovirus expression system (Life Technologies). Cells were harvested by centrifugation and lysed by Dounce homogenizer in 50 mM HEPES (pH 7.4), 300 mM NaCl, 30 mM imidazole, 5% glycerol, 5 μg/mL DNase, 0.5% Triton X-100, 0.5 mM TCEP, 1 mM PMSF, and protease inhibitor cocktail.
Following incubation in the dark for 2 hrs at 37 º C , the labeled protein was separated from the solution by size exclusion chromatography using PD MiniTrap G25 (GE Healthcare).  supplemented with 10% fetal bovine serum (Invitrogen) and 1% PenStrep-Glutamine in a humidified incubator with 5% CO2 at 37℃. CRISPR-cas9 gene editing system was used to generate the PLC1 knockout Jurkat T cell line.
Jurkat T cell lines stably expressing a fluorescence reporter was generated by lentiviral transduction. HEK293T cells were transfected with the pHR plasmids encoding the gene of interests, and the viral packaging plasmids pMD2.G and psPAX2 using Genejuice (EMD millipore). 72 hr after plasmid transfection, cell culture media containing viral particles were harvested and added into Jurkat cells for infection in RPMI media overnight.
Jurkat cells expressing fluorescent proteins were sorted by FACS to generate a stable and homogenous expression population.

Live cell imaging of microcluster formation in Jurkat T cells
A 96-well glass-bottom imaging plate was coated with 5 μg/mL anti-CD3 antibody (OKT3) in PBS overnight at room temperature. Unbound OKT3 was washed out the next morning.
The imaging plate was equilibrated with image medium (RPMI Medium 1640 without phenol red, 20 mM HEPES, pH 7.4). 0.1 million of Jurkat T cells were added to each well and microclusters were imaged by TIRF microscopy.

Immunoblot analysis
Jurkat T cells were washed with PBS three times, incubated at 37℃ for 30 min, and stimulated with 2 μg/mL OKT3 (eBioscience #16-0037-85) and 2 μg/mL anti-CD28 antibody (eBioscience #16-0289-85). The reaction was stopped by directly adding 2x SDS-PAGE loading buffer (Bio-Rad #1610737) containing protease inhibitor cocktail (Roche #11873580001). The lysates were boiled for 10 min at 95℃ and clarified by centrifugation at 12, 000 rpm for 15 min at 4℃. The supernatants were loaded onto a 4-20% protein gel (Bio-Rad #4568096) for SDS-PAGE analysis, followed by a transferring onto PVDF membrane (Bio-Rad #1620177). The membrane was blocked with TBST buffer containing 5% nonfat milk for 1 hr at room temperature, and blotted with indicated primary antibodies overnight at 4℃. The next day, the membrane was further blotted with HRP-conjugated secondary antibody for one hour at room temperature. Target proteins were detected with a chemiluminescent HRP substrate (Thermo Scientific #34577) and visualized by a Bio-Rad ChemiDoc imaging system.

Data analysis
Images were analyzed using FIJI (Image J). LAT clustering was quantified as normalized variance, which equals to the square of standard deviation divided by mean (after subtracting background). Graphs in the same panel were displayed with the same brightness and contrast setting. The half recovery time of FRAP was obtained by fitting a "one-phase association" model in GraphPad Prism 7.00 software.

Computer simulations
The computational model consists of two-dimensional particles of circular shape and diameter σ, with patches playing the role of binding sites ( Figure 4A). All particles interact with each other through Weeks-Chandler-Anderson potential, of strength 10 kBT, 2d molecular dynamics simulations are performed using LAMMPS (Plimpton, 1995).
There, patches are implemented by means of ghost particles of diameter equal to 0.05 σ and positioned at fixed distance 0.475 σ from the center of the proper particle. Two interacting patches (as per Figure 4A) interact via a cosine-squared potential with range of attractiveness 0.15 σ between the centers of the patches. The maximum depths ϵ of the cosine-squared potential is set to 30 kBT, kB being the Boltzmann constant and T being the temperature, rendering the bonds practically unbreakable within simulation time. Each molecule, i.e. a volume-excluded particle together with its patches representing binding sites, is treated as a rigid body. The angular distance between patches belonging to the same molecule is chosen in such a way to 1) make binding between two patches exclusive, and 2) allow a Sos1 and a Grb2 molecule to bind through either one or two pairs of patches. 2d Brownian dynamics is ensured by an overdamping Langevin thermostat, enforcing room temperature at every step.
In simulations, the number of LAT molecules is fixed to 200. The ratio of LAT:Sos1:Grb2 particles is kept constant at 1:1:2, whereas the proportion PLC1:LAT is varied from 0 to 3. Our goal is not to capture the exact experimental value of these ratios, both because in experiments the surface density of cytosolic cross-linkers at the membrane is unknown and because it can only affect quantitatively, but not qualitatively, the results. Different surface densities ρLAT are chosen for LAT, from 0.005 to 0.05 σ -2 , with no noticeable difference in the mechanism of clustering (see Figure S4A and SI). In the text we show results for ρLAT = 0.02 σ -2 .
The coalescence likelihood is defined as the number of possible bonds that can form, in principle, between a cluster and another one, summed over all clusters available at a given time. This amounts to where na,i is the total number of yet unbound binding sites of type i present in cluster a, the first sum runs over all combinations of two clusters, labeled a and b, and the second sum runs over all types of interacting patch pairs (i, j), as per Figure 4A. For simplicity, spatial hindrance is neglected in the computation of the coalescence likelihood: all unbound binding sites are counted, irrespective of whether they are physically accessible or not. This assumption is justified by the fact that for clusters of the size and structure that we observe, the vast majority of molecules is situated at the boundary and can therefore contribute to cluster coalescence.
The compactness parameter mentioned in the text is based on the analysis of the gyration tensor of each cluster, using as pole the cluster's center of mass (the gyration tensor is equal to the inertia tensor for unit masses). The tensor is diagonalized, in order to obtain the three principal moments of inertia: one of them, Iz, represents the inertia of in-plane rotations about the axis z, perpendicular to the plane and going through the center of mass; the remaining two principal moments I1 and I2 represent the squares of the gyration radii of the cluster along two perpendicular directions in the plane. It is possible to compute the moment of inertia Iz,min along z of an ellipse of gyration radii √ 1 and √ 2 , made of close-packed circles (continuous bulk density for hexagonal close-packing is assumed): this represents the inertia of a fictional cluster of approximately the same shape as the real one, made of as many circular particles as there can fit. The compactness parameter is then defined as the ratio Iz,min / Iz and tends to 0 for a maximally sparse cluster and to 1 for a maximal-mass-density ellipse with no void regions (see Figure 4B for examples and Section SI 4).     (B) Recombinant proteins used in this study. Purified proteins were applied to SDS-PAGE, followed by Coomassie blue staining.
Movie S2. LAT cluster formation with Grb2, Sos1 and PLC1. Same condition as in Movie S1 except that PLC1 (50 nM) was added together with Grb2 and Sos1 at 0 sec.

Movie S3.
Early phase of simulation of LAT cluster formation at low PLC1 to LAT ratio.
The simulation involves 200 LAT, 15 PLC1, 400 Grb2, and 200 Sos1 molecules, all in a monomeric state. The simulation starts at second 2.00, corresponding to timestep 0, and ends at timestep 10x10 6 . The interval between two frames is 0.2x10 6 timesteps and the frame rate is 8 s -1 . Particles scheme is the same as in Figure 4A: grey-LAT, yellow-PLC1, blue-Grb2, and pink-Sos1.
Movie S4. Full-length simulation of LAT cluster formation at low PLC1 to LAT ratio (high resolution movie for visualizing individual chemical bonds). The simulation involves 200 LAT, 15 PLC1, 400 Grb2, and 200 Sos1 molecules, all in a monomeric state. The simulation starts at second 2.00, corresponding to timestep 0, and ends at timestep 50x10 6 . The interval between two frames is 0.2x10 6 timesteps and the frame rate is 16 s -1 . Particles scheme is the same as in Figure 4A: grey-LAT, yellow-PLC1, blue-Grb2, and pink-Sos1. molecules, all in a monomeric state. The simulation starts at second 2.00, corresponding to timestep 0, and ends at timestep 10x10 6 . The interval between two frames is 0.2x10 6 timesteps and the frame rate is 8 s -1 . Particles scheme is the same as in Figure 4A: grey-LAT, yellow-PLC1, blue-Grb2, and pink-Sos1. state. The simulation starts at second 2.00, corresponding to timestep 0, and ends at timestep 50x10 6 . The interval between two frames is 0.2x10 6 timesteps and the frame rate is 16 s -1 . Particles scheme is the same as in Figure 4A: grey-LAT, yellow-PLC1, blue-Grb2, and pink-Sos1.

Movie
Movie S7. Early phase of simulation of LAT cluster formation at high PLC1 to LAT ratio.
The simulation involves 200 LAT, 600 PLC1, 400 Grb2, and 200 Sos1 molecules, all in a monomeric state. The simulation starts at second 2.00, corresponding to timestep 0, and ends at timestep 10x10 6 . The interval between two frames is 0.2x10 6 timesteps and the frame rate is 8 s -1 . Particles scheme is the same as in Figure 4A: grey-LAT, yellow-PLC1, blue-Grb2, and pink-Sos1.  Step 1

Movie
Step 2

Disassociation
Step 4 Bonds and clusters A bond between two molecules is defined by two interacting patches being within range of attraction r c . Two molecules are in the same cluster if and only if they are connected by a path of molecules, such that each molecule forms a bond (as just defined) with the next one. Given the short range and the strength of the interaction, which makes bonds irreversible, clusters are robust against reasonably small variations of the cutoff distance used to define a bond. All the software used in the analysis of simulation results, including clustering, was custom-made.

SI 2 Kinetics
To study the kinetics of clusters formation, and later the composition of clusters, we compute two quantities: average cluster size and average coordination. The average cluster size is defined by the number of LAT molecules present on average in a cluster at a given simulation time, for a given point in the parameters space. This corresponds to the total number of LAT molecules divided by the total number of clusters, averaged over different realizations. The average coordination, for LAT, PLCγ1, Sos1, or Grb2 molecules, is defined as the number of bonds formed on average by a molecule of that type (i.e. the number of its occupied binding sites). Fig. SI 1 shows the time evolution of these two quantities. The average coordination (Fig. SI 1B) exhibits a quick relaxation to its long-time value. This is because most of the bonds are formed during the first 10 7 timesteps of the simulations. Later on, bonds become rarer, because many binding sites are already occupied, and because the probability of two particles (or clusters) hitting each other by thermal motion decreases as particles condense locally.
On the other hand, the average cluster size (Fig. SI  1A) is dramatically affected by these late-time binding events, as clusters typically double their size upon binding with one another. At low to medium PLCγ1 concentrations, the curves will ideally grow until all particles are condensed into one big cluster: this is a consequence of irreversible interactions and, ultimately, finite size. In a real system, where bonds can break, the equilibrium cluster size will be set by the balance between bonds dynamics (how often bonds break) and density/entropy (how often clusters bump into each other and new bonds form). On the contrary, at high PLCγ1 concentrations, the linkers saturation phenomenon soon forces the system into a locked state, where further clustering is blocked by the unavailability of binding sites.
An interesting feature of Fig. SI 1A is that plots for different ρ LAT exhibit the same behavior. Surface density seems to act simply as a rescaling parameter and to influence only the rate of cluster formation: the smaller the density, the less frequently clusters will hit each other risking to coalesce. Analogously, in a real system at equilibrium, the smaller the density, the higher will be the entropic cost associated to bond formation.
Even though our simulations are not meant to reproduce quantitative results, but to give a physically sound explanation to observed experimental phenomena, it can be argued that our model represents realistically an early stage of clustering, where the number of particles is low and bonds have not had the time to break yet. Nonetheless, the non-monotonicity of the average cluster size is a persistent feature, present at all simulated times (except obviously very early stages, where most bonds are not formed yet) and at all densities (Figs. SI 2A and S4A). This suggests that the mechanism behind this re-entrant behavior is robust and at least qualitatively independent from time and ρ LAT . Motivated by this observation, in the following (as in the main text) we present in detail data captured at time t 0 = 0.5 × 10 8 steps (when needed, for the sake of conciseness, we also restrict ourselves to a density of ρ LAT = 0.02 σ −2 ). Our conclusions reasonably hold true, irrespective of this particular choice. As previously observed, these curves do not depend on density, except for a slight tendency to decrease as density decreases: this is, again, because at equal time, a lower density system has experienced less collisions than a high density one, and this reflects on the dynamics of the coordination curves (see Fig. SI 2B). To help interpret what happens, we focus on ρ LAT = 0.02 σ −2 and break down the four curves from Fig. SI 3, according the the weight of each type of bond. This is done in Fig. S4B, that prompts the following observations.

SI 3 Coordination and coalescence
-The PLCγ1 cSH2 domain compete with the Grb2 SH2 domain for the pY171 site on LAT (top left plot, light orange vs dark blue bar).
-As PLCγ1 concentration increases, LAT's pY132 and pY171 become completely saturated, due to excess of SH2 binding sites from PLCγ1. Remaining pY191 and pY226 are slightly affected: competition with PLCγ1 for pY171 redirect Grb2's SH2 to other LAT sites (medium and light blue bars on gray background), so that overall Grb2 binding to LAT stays constant (gray band in bottom right plot).
-At high PLCγ1 concentrations, Sos1 is completely saturated (bottom left plot), due to the overwhelming abundance of SH3 sites from PLCγ1 (yellow bars). In addition, this SH3 domain on PLCγ1 competes for Sos1 with its homologues on Grb2 (light blue bars on pink background), thus causing a decrease in the overall average coordination of Grb2 (bottom right plot).
-Although its presence favors high coordination of LAT and Sos1, and increases in absolute value the number of bonds, PLCγ1's average coordination decreases with concentration (top right plot). This is due to scarcity of boundable sites on LAT (gray band), and, to a minor extent, of Sos1 (pink band).
All this provides a straightforward interpretation for Fig. 4C, where the coalescence likelihood is observed to decrease drastically, as Sos1 and LAT saturate. The main effect is due to Sos1, which with its 12 binding combinations (Fig. 4A) is involved in the majority of bonds formed in the system.

SI 4 Shape and compactness
In relation to FRAP experiments, we further analyzed the structure of the networks formed within each cluster, as alluded to in Fig. 4D and in the Methods section. We performed first an analysis of the moments of gyration (or inertia) based on standard Euclidean metric, and then a graph-theoretical analysis based on the connections between molecules rather than on their positions. The latter was motivated by the fact that the geometrical features of our model mainly result from the need to impose bond specificity and exclusivity, and are therefore not necessarily physical. The two analyses are however consistent with each other, showing, as expected, a certain degree of correlation between geometry and network structure.
The mechanical analysis was performed by computing inertia tensors of clusters, relative to their center of mass. To this purpose, since the property of interest was of geometrical and not dynamical origin, all molecules were assigned a unit mass. The center of mass was computed by accounting for periodical boundary conditions during the procedure of clustering analysis. As described in the Methods section, the Coordination is defined as the number of bound sites, so it ranges from 0 to 4 for LAT and Sos1, and from 0 to 3 for PLCγ1 and Grb2. eigenvalues of the inertia tensor I z , I 1 and I 2 were determined. I 1 and I 2 , the squares of the two in-plane gyration radii, were used to quantify two-dimensional shape through a parameter that we called Roundness and defined as follows: with I 1 < I 2 . Taking this ratio amounts to approximating each cluster with an ellipse and estimating its eccentricity. The Roundness tends to 0 for clusters in the shape of a straight line and to 1 for in-planerotation symmetric clusters, such as circular ones. We then compared the in-plane-rotation principal moment of inertia (I z ) to the same quantity (I z,min ) for a homogeneous ellipse of equal roundness, with a density equal to the bulk density of close-packed circles. The latter is the maximum possible density and gives rise to the minimum possible inertia. In practice, for a given cluster, I z,min = N 2 σ 2 16η where η = π/(2 √ 3) is the packing fraction of a hexagonal close-packed lattice of circles and N is the number of molecules in the cluster. This defines the following Compactness parameter: This Compactness intuitively quantifies the presence of holes in the cluster structure: it is 1 if particles are maximally packed (this is impossible in our case, due to the geometry of bonds, so that the achievable Compactness limit is actually below 1), whereas it tends to 0 for a sparse cluster, with many void regions. Roundness and Compactness, averaged over clusters mixed up from all different realizations, are shown in Fig. SI 4. Once again, ρ LAT does not seem to play a role. An effect of PLCγ1 concentration on the Roundness is not recognizable: this would require a symmetry breaking phenomenon that does not seem plausible. On the other hand, Compactness increases with PLCγ1 concentration, as observed in the main text.
The graph-theoretical analysis was performed by defining for each cluster an equivalent undirected unweighted graph, where nodes correspond to particles and edges connect two nodes whose corresponding particles are linked by at least one bond. We characterized the sparsity of such graphs by computing the ratio between number of edges |E| and number of nodes |V | = N . Indeed, a sparse graph will exhibit a linear proportionality between these two quantities, whereas for a fully connected graph |E| ∼ |V | 2 . Surprisingly, although this ratio cannot exceed 2 in our system because of limited particle valence, it does not show any appreciable increase with PLCγ1 concentration. This is in apparent contradiction with the fact that clusters appear more compact as PLCγ1 is added. The reason is that, due to saturation, most of the added PLCγ1 molecules form only one bond, as confirmed by Figs. 4E and S4B, thus increasing the number of nodes with just one edge (terminal nodes or 'leaves'). This effect is compensated by an increase in the coordination of LAT and Sos1 (see again Figs. 4E and S4B). The result is a constant sparsity. This observation is coherent with and reinforces our picture of available-bonds-limited cluster growth, symbolized by a decreasing coalescence likelihood. Finally, we attempted to provide a graph-theoretical equivalent to the Compactness parameter defined earlier on a mechanical basis. We defined a graphtheoretical moment of inertia about the center of mass (I z,GT ): this exploits on the one hand the concept of graph-theoretical distance d (u, v), i.e. the number of edges forming the shortest possible continuous path between nodes u and v, and on the other hand the fact that the moment of inertia relative to the center of mass is the smallest one possible (a corollary of the Huygens-Steiner theorem). The graph-theoretical moment of inertia is given by where V is the set of nodes. This needs to be compared to a similar quantity for a compact cluster; since the latter will scale as |V | 2 , for simplicity we define the graph-theoretical compactness as to be compared to Eq. (3). The average of this quantity is plotted in Fig. SI 5 (blue) and increases with PLCγ1 concentration, as expected. Another measure of the compactness of our graphs originates from the comparison between the diameter D of a graph (the distance between the two most distant nodes) and its theoretical lower bound, approximately given by the Moore limit for vertices of maximum degree 4: The average of the quantity is shown in green in